Tech

Cerebras hit $60B at IPO, but it nearly burned out chasing a wafer-scale chip

Julie Bort ·
Cerebras hit $60B at IPO, but it nearly burned out chasing a wafer-scale chip

Cerebras Systems ended its IPO week with a market value near $60 billion, an outcome that lifted its co-founders into the billionaires club and confirmed Wall Street's appetite for purpose-built AI silicon. Behind the celebration, however, is a story the company is only now telling in detail: in 2019 Cerebras came close to running out of cash while trying to ship a chip the rest of the semiconductor industry believed could not be built.

In a wide-ranging interview, CEO Andrew Feldman said the startup spent roughly $200 million and burned about $8 million a month tackling a single problem, packaging a processor that was 58 times larger than a standard chip and consumed about 40 times the power. Cooling, power delivery, and getting data on and off the wafer all had to be reinvented. Engineers even built custom rigs to drive 40 screws simultaneously without cracking the silicon.

The breakthrough finally came in July 2019, when a properly packaged wafer-scale unit worked end to end. That part now powers inference workloads at OpenAI and Amazon Web Services and underpins the bullish thesis investors took into last week's listing.

Why the IPO matters beyond Cerebras

The debut is a meaningful data point for the broader AI hardware market, where investors have been searching for credible challengers to Nvidia. Cerebras's pitch is that wafer-scale designs are better suited to large-model inference than racks of discrete GPUs, and customers like OpenAI, which lent the company about $1 billion secured by warrants, are voting with their workloads.

The disclosure of the near-death period is also a reminder that hardware startups remain capital-intensive and unforgiving. Software-style pivots are not available when the bet is a single die the size of a dinner plate.

aisemiconductorscerebrasipoopenai
Original source
TechCrunch →

Related articles

Tech Now
Tech

Tech Now

Shiona McCallum meets the medical expert testing out the future of prosthetics following his own near-fatal cycling accident. Paul Carter heads inside Amazon

Tech Life
Tech

Tech Life

We talk to an expert about the risks from AI hackers

Tech Life
Tech

Tech Life

What is the role of a teacher once AI arrives in the school classroom?